Should your hotel assistant admit it's an AI?

Should your hotel's AI assistant tell guests it is an AI? Two properties we work with answered in opposite ways, and both work.

One introduces its assistant by name as a digital agent and says a human team is available. The other gives the assistant a name and a character and never calls it digital. Each works, because each is consistent.

That is the part most people get wrong. They treat disclosure as a moral question with one right answer, argue about it for a week, and then ship an assistant that is openly digital on the website, anonymous on WhatsApp and signed by a made up person in the pre arrival email. The guest notices the seam long before they notice the policy.

What research says

There is now a study worth reading before you decide. Research led by Babak Taheri at Texas A&M, published in the International Journal of Hospitality Management and written up by Texas A&M AgriLife in May 2026, surveyed 340 adults in the United Kingdom who had used chatbots to book hotels. Read 28 September 2026.

Three things about it matter to you.

First, what actually produced discomfort. The researchers looked at three chatbot flaws: inaccuracy or unreliability, deceptiveness, and intrusiveness. All three increased discomfort, but inaccuracy was by far the strongest, with a path coefficient more than four times larger than the next factor. An assistant that gives wrong rates or misreads a cancellation policy unsettles guests more than one that is coy about being a bot.

Second, what discomfort cost. In the survey, it reduced users' willingness to keep interacting with the chatbot by nearly 38 percent, and roughly doubled the likelihood that they would delay or abandon a booking decision.

Third, the human likeness point, stated as the researchers state it. They describe an uncanny valley effect, where a chatbot's failures become more disturbing the more closely it mimics human speech. Taheri's explanation in the write up is that users implicitly expect a human like system to behave like a human, and the gap between expectation and reality triggers something deeper than disappointment. The researchers recommend that hotel platforms open conversations with a clear statement such as "Hi, I'm your AI assistant", and give users an easy way to reach a human for complex questions.

Note what this does and does not say. It is one survey, in one country, of people recalling their own experiences. It does not prove that a named character always performs worse than an open disclosure. What it does support is narrower and more useful: the more human your assistant sounds, the more expensive its mistakes become, and accuracy is the thing guests punish hardest.

The three choices

1. Named and openly digital. The safest, and the one closest to what the researchers recommend. It sets expectations about tone and, more importantly, about what the assistant can approve. A guest who knows they are talking to software does not assume a discount has been authorised.
2. Named with a character, not labeled. Works for playful brands, on one condition: the assistant never claims to be a person when asked. A character is a voice, not a lie.
3. No name, plain "the hotel". Fine for formal properties, business hotels and anywhere the brand speaks in the first person plural anyway. It is also the easiest to keep consistent across four languages.

If you cannot decide, take the first. The cost of being open is a small amount of charm. The cost of being caught is trust, and the study suggests you pay it at exactly the moment the guest was about to book.

Rules that hold in every case

Whichever option you pick, three rules do not bend.

• If a guest asks "am I talking to a bot?", the answer is yes, always. Immediately, plainly, without a joke that dodges the question. This is the one that decides whether the character in option two is charm or deception.
• The assistant never claims to have seen, checked or done something a person did. "I've checked with housekeeping" is a lie if nobody was asked. "I've sent this to housekeeping and they'll confirm" is true and just as reassuring. The difference takes one word and it is the difference between a persona and a fabrication.
• The hand off to a human is always visible. The guest should be able to reach a person without arguing with the assistant, and should be told when it happens. The Texas A&M write up makes the same recommendation, and it is also the thing your team will thank you for. Where you draw that line is a business decision, not a technical one, and we work through it in human in the loop versus fully automated guest messaging.

Add a fourth if you send proactive messages: be as clear about who is writing in a template as you are in a live chat. Consistency across webchat, WhatsApp and templates is the whole point.

Takeaway

Pick one persona, write it down, and keep it consistent across webchat, WhatsApp and templates.

Then spend the rest of your effort on accuracy, because that is what the research says guests actually react to. That means the assistant has to be fed by a knowledge base that somebody owns and keeps current, and it has to know when to stop and hand the conversation to a person. A persona sitting on top of wrong information is not a brand decision. It is a complaint waiting for a date.